LLMBridge

Test copy of the proposal on semantic-mediawiki.org, used to test Template:AgendaDay26. Time slot and status are test values from the draft program.

LLMBridge (Lightning talk, 5 min, Remote, Accepted)

Speakers: Shreenivas Madagundi · Start: 2026/11/12 14:30:00 · Conference: MediaWiki Users and Developers Conference Fall 2026

LLMBridge uses Google Gemini 2.5 Flash to analyze page images and generate keywords that improve content discoverability and help search systems better understand wiki pages. MediaWiki pages often contain valuable information embedded in images, diagrams, screenshots, and flowcharts. Traditional search indexing may not fully understand this visual content, which can reduce page discoverability.

LLMBridge bridges this gap by integrating MediaWiki with Google Gemini 2.5 Flash through Vertex AI. The extension analyzes images associated with wiki pages and extracts descriptive keywords representing their content, objects, colors, mood, and style. The generated keywords can be stored as page metadata or used by downstream search and indexing processes. This provides additional context for search systems, helping improve relevance, content discovery, and page ranking.

The extension supports MediaWiki file repositories and common image formats such as PNG, JPEG. It is designed with reusable services so organizations can integrate AI-powered image analysis into maintenance scripts, upload workflows, jobs, and other MediaWiki features.

The presentation will cover:

- The motivation for enriching MediaWiki search metadata - LLMBridge architecture and MediaWiki integration - Vertex AI authentication and Gemini 2.5 Flash requests - Image retrieval and MIME-type handling - Keyword extraction and normalization - Integration with page indexing and search workflows - Operational considerations, including cost, privacy, errors, and validation - Future possibilities such as OCR, summaries, alt text, and multimodal search LLMBridge does not guarantee higher rankings by itself. Instead, it enriches page metadata and gives search systems better information to evaluate relevance.